资源环境科技发展态势分析平台

In this paper, we propose an ensemble optimization algorithm to retrieve the hydrometeor profiles for the upcoming Ice Cloud Imager submillimeter-wave radiometer. The algorithm combines the database approach with local optimization under a framework of Bayesian Monte Carlo methodology. It first builds a retrieval database with atmospheric parameters obtained from prior information, and then it conducts ensemble optimization to minimize the cost function when too few database cases match the observations. The ensemble approach does not use the gradient information. Instead, it learns what the optimum's posterior distribution looks like, and then tries to estimate this distribution directly. The algorithm represents the unknown continuous posterior distribution with an ensemble of discrete cases, and it iteratively decreases the estimate uncertainties by generating a new ensemble from the learned distribution. Once the termination criterion is met, retrieval results and uncertainties are derived by integration over the final ensemble. We conduct retrieval experiments using simulated noisy brightness temperatures. Retrieved profiles are compared to the true values, and the results are statistically analyzed. The ensemble optimization algorithm is demonstrated to be an effective objective value minimization technique that avoids the complexity of gradient radiative transfer calculations. The algorithm also shows a satisfactory water content profile retrieval performance for frozen hydrometeors.